Pricing

Free to use. Pay for what you run.

One rate card, shared by MindsHub Cowork and the Unified Inference API. No subscription, no plan to choose.

Free MindsHub Air, 5,000,000 tokens a month. No card.
Your keys Any model on your own provider accounts — you pay them, not us.
Credits Everything else, at the rates below plus a 5% platform fee.

Rate card

Published rates, per model.

What credits buy, per model. The same rates bill Cowork and the Unified Inference API. Expand any row for its full billing rules — long-prompt tiers, web search, cache-write waivers, reasoning levels and failover order.

Model	Provider	Input	Output	Cached input	Cache write
Our alias, kept on whichever model balances best Free allowance: the first 5M Air tokens each month are on us	MindsHub	$0.20	$1.20	$0.02	$0.25
Free allowance 5M tokens a month at no charge, in Cowork and through the API alike. The rates here apply beyond the allowance. Air is an alias we keep pointed at the model that strikes the best balance of cost, speed and intelligence, so both the target and these rates can move as better options ship. Long prompts — 272K+ tokens $0.40 input $1.80 output $0.04 cached input $0.50 cache write Per 1M tokens. Once a prompt crosses 272K tokens, these rates replace the standard ones for the whole request — the same way MindsHub bills it. Failover order Gemini 3.7 Flash Tried in this order if MindsHub Air is unavailable, so your work keeps moving.
Balanced everyday workhorse	Anthropic	$2.00	$10.00	$0.20	$2.50
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $10.00 per 1,000 searches Anthropic's own search, built into the model. Reasoning effort `lowmediumhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order GPT 5.6 Sol Tried in this order if Claude Sonnet 5 is unavailable, so your work keeps moving.
Tops the intelligence index; built for long agentic runs	Anthropic	$5.00	$25.00	$0.50	$6.25
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $10.00 per 1,000 searches Anthropic's own search, built into the model. Reasoning effort `lowmediumhighxhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order GPT 5.6 Sol Tried in this order if Claude Opus 5 is unavailable, so your work keeps moving.
Anthropic's top tier for the hardest problems	Anthropic	$10.00	$50.00	$1.00	$12.50
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $10.00 per 1,000 searches Anthropic's own search, built into the model. Reasoning effort `lowmediumhighxhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order GPT 5.6 Sol Tried in this order if Claude Fable 5 is unavailable, so your work keeps moving.
Fastest Claude, for high-volume simple work	Anthropic	$1.00	$5.00	$0.10	$1.25
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $10.00 per 1,000 searches Anthropic's own search, built into the model. Failover order Gemini 3.7 Flash DeepSeek V4-Pro-0813 Tried in this order if Claude Haiku 4.5 is unavailable, so your work keeps moving.
Flagship all-rounder, and first on the coding index	OpenAI	$5.00	$30.00	$0.50	$6.25
Long prompts — 272K+ tokens $10.00 input $45.00 output $1.00 cached input $12.50 cache write Per 1M tokens. Once a prompt crosses 272K tokens, these rates replace the standard ones for the whole request — the same way OpenAI bills it. Web search $10.00 per 1,000 searches OpenAI's own search, built into the model. Reasoning effort `nonelowmediumhighxhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send medium unless you ask for another. Failover order Claude Opus 5 Gemini 3.1 Pro Preview Tried in this order if GPT 5.6 Sol is unavailable, so your work keeps moving.
Balanced mid-tier for everyday work	OpenAI	$2.00	$12.00	$0.20	$2.50
Long prompts — 272K+ tokens $4.00 input $18.00 output $0.40 cached input $5.00 cache write Per 1M tokens. Once a prompt crosses 272K tokens, these rates replace the standard ones for the whole request — the same way OpenAI bills it. Web search $10.00 per 1,000 searches OpenAI's own search, built into the model. Reasoning effort `nonelowmediumhighxhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send medium unless you ask for another. Failover order Claude Sonnet 5 Gemini 3.1 Pro Preview Tried in this order if GPT 5.6 Terra is unavailable, so your work keeps moving.
Fast and cheap for high-volume work	OpenAI	$0.20	$1.20	$0.02	$0.25
Long prompts — 272K+ tokens $0.40 input $1.80 output $0.04 cached input $0.50 cache write Per 1M tokens. Once a prompt crosses 272K tokens, these rates replace the standard ones for the whole request — the same way OpenAI bills it. Web search $10.00 per 1,000 searches OpenAI's own search, built into the model. Reasoning effort `nonelowmediumhighxhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send medium unless you ask for another. Failover order Gemini 3.7 Flash Claude Haiku 4.5 Tried in this order if GPT 5.6 Luna is unavailable, so your work keeps moving.
Coding specialist for repo-scale changes	OpenAI	$1.75	$14.00	$0.18	$1.75
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $10.00 per 1,000 searches OpenAI's own search, built into the model. Reasoning effort `lowmediumhighxhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send medium unless you ask for another. Failover order Claude Sonnet 5 Tried in this order if GPT 5.3 Codex is unavailable, so your work keeps moving.
Small and cheap for routine automation	OpenAI	$0.75	$4.50	$0.08	$0.75
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $10.00 per 1,000 searches OpenAI's own search, built into the model. Reasoning effort `nonelowmediumhighxhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send none unless you ask for another. Failover order Gemini 3.7 Flash Claude Haiku 4.5 Tried in this order if GPT 5.4 Mini is unavailable, so your work keeps moving.
Built for throughput on bulk, simple work	OpenAI	$0.20	$1.25	$0.02	$0.20
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $10.00 per 1,000 searches OpenAI's own search, built into the model. Reasoning effort `nonelowmediumhighxhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send none unless you ask for another. Failover order Gemini 3.7 Flash Tried in this order if GPT 5.4 Nano is unavailable, so your work keeps moving.
Huge documents, images, audio and video	Google	$2.00	$12.00	$0.20	Free
Long prompts — 200K+ tokens $4.00 input $18.00 output $0.40 cached input Per 1M tokens. Once a prompt crosses 200K tokens, these rates replace the standard ones for the whole request — the same way Google bills it. Web search $14.00 per 1,000 searches Google's own search, built into the model. Cache writes Free on this model — you're billed only when cached content is read back, at the cached-input rate above. Reasoning effort `lowmediumhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order Claude Sonnet 5 GPT 5.6 Sol Tried in this order if Gemini 3.1 Pro Preview is unavailable, so your work keeps moving.
Fast, high-volume everyday tasks	Google	$0.75	$3.75	$0.08	Free
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $14.00 per 1,000 searches Google's own search, built into the model. Cache writes Free on this model — you're billed only when cached content is read back, at the cached-input rate above. Reasoning effort `lowmediumhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send medium unless you ask for another. Failover order GPT 5.4 Mini DeepSeek V4-Pro-0813 Tried in this order if Gemini 3.7 Flash is unavailable, so your work keeps moving.
	Google	$0.75	$3.75	$0.08	Free
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $14.00 per 1,000 searches Google's own search, built into the model. Cache writes Free on this model — you're billed only when cached content is read back, at the cached-input rate above. Reasoning effort `minimallowmediumhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send medium unless you ask for another. Failover order GPT 5.4 Mini DeepSeek V4-Pro-0813 Tried in this order if Gemini 3.6 Flash is unavailable, so your work keeps moving.
Long autonomous coding runs you can self-host	Fireworks AI	$3.00	$15.00	$0.30	$3.00
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Failover order DeepSeek V4-Pro-0813 Tried in this order if Kimi K3 is unavailable, so your work keeps moving.
Open-weight reasoning and coding at a budget rate	Fireworks AI	$1.32	$3.96	$0.05	$1.32
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Reasoning effort `lowhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order Kimi K3 Tried in this order if DeepSeek V4-Pro-0813 is unavailable, so your work keeps moving.
	Fireworks AI	$1.74	$3.48	$0.15	$1.74
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Reasoning effort `lowhighmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order Kimi K3 Tried in this order if DeepSeek V4-Pro-0813 is unavailable, so your work keeps moving.
Multilingual coverage at a fraction of the flagship rate	Fireworks AI	$2.00	$6.00	$0.25	$2.00
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Reasoning effort `lowmediumxhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send xhigh unless you ask for another. Failover order DeepSeek V4-Pro-0813 Tried in this order if Qwen3.8-2.4T-A95B is unavailable, so your work keeps moving.
	Fireworks AI	$0.40	$1.60	$0.08	$0.40
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Failover order DeepSeek V4-Pro-0813 Tried in this order if Qwen3.7 Plus is unavailable, so your work keeps moving.
Open weights with the GLM coding profile	Fireworks AI	$1.40	$4.40	$0.14	$1.40
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Reasoning effort `highmax` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send max unless you ask for another. Failover order DeepSeek V4-Pro-0813 Tried in this order if GLM 5.2 is unavailable, so your work keeps moving.
Every input type, at an open-model rate	Meta	$1.25	$4.25	$0.15	$1.25
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Failover order Claude Sonnet 5 Kimi K3 Tried in this order if Muse Spark 1.2 is unavailable, so your work keeps moving.
	Meta	$1.25	$4.25	$0.15	$1.25
Long prompts No surcharge — the full context window bills at the standard rates above. Web search $7.00 per 1,000 searches $1.00 per 1,000 page fetches Served by Exa, which bills page fetches separately. Failover order Claude Sonnet 5 Kimi K3 Tried in this order if Muse Spark 1.1 is unavailable, so your work keeps moving.
Reasoning with access to the live web	SpaceXAI	$2.00	$6.00	$0.50	$2.00
Long prompts — 200K+ tokens $4.00 input $12.00 output $1.00 cached input $4.00 cache write Per 1M tokens. Once a prompt crosses 200K tokens, these rates replace the standard ones for the whole request — the same way SpaceXAI bills it. Web search $5.00 per 1,000 searches SpaceXAI's own search, built into the model. Reasoning effort `lowmediumhighxhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order Claude Sonnet 5 Kimi K3 Tried in this order if Grok 4.6 is unavailable, so your work keeps moving.
The previous Grok, and now the better-value one	SpaceXAI	$2.00	$6.00	$0.30	$2.00
Long prompts — 200K+ tokens $4.00 input $12.00 output $0.60 cached input $4.00 cache write Per 1M tokens. Once a prompt crosses 200K tokens, these rates replace the standard ones for the whole request — the same way SpaceXAI bills it. Web search $5.00 per 1,000 searches SpaceXAI's own search, built into the model. Reasoning effort `lowmediumhigh` Dial thinking up or down per task. Higher effort spends more output tokens, billed at the same rate. We send high unless you ask for another. Failover order Claude Sonnet 5 Kimi K3 Tried in this order if Grok 4.5 is unavailable, so your work keeps moving.

Token prices are USD per 1M tokens; web search and page fetches are USD per 1,000 calls. Cached input is the discounted rate for prompt content the provider already has cached, and cache write is what it costs to put it there. Expand any model for its long-prompt rates, search pricing and failover order. We mirror each provider's own billing rules rather than inventing our own, and a 5% platform fee is added to your bill on top of these rates.

Deciding between models? The ranking board compares the whole catalog on quality, speed and measured cost per finished task, alongside the models coming to it next.

MindsHub Foundry

Double your first 3 months of credit.

A limited cohort of teams works directly with our product team, gets early access to what we build, and receives a 100% match on their first 3 months of inference-credit purchases — up to $5,000. Applications are reviewed on a rolling basis.

About the Foundry

Cowork runs in your browser and in the desktop app — the same workspace either way. Get the app for macOS or Windows.

Support follows the balance rather than a plan: the community Discord while you are on the free allowance, and a tracked support ticket once you have added credits.

FAQ

Common questions.

What does it cost to start?

Nothing. Every account gets MindsHub Air with 5,000,000 tokens a month at no charge — no card, and no plan to choose. Input, output, cached reads and cache writes all count toward the same allowance, and it refreshes at the start of each month.

How do I use a model other than MindsHub Air?

Add credits. Usage then draws down that prepaid balance at the published per-model rates on this page, plus a 5% platform fee. Credits reach the whole catalog, served through the MindsHub Model Router with automatic failover, so you hold no provider accounts. Add credit when you want, or turn on auto-recharge with a cap on how much it can charge each month.

Can I use my own provider API keys instead?

Yes. Paste in keys you already have and MindsHub calls those providers directly — you pay the provider, not us, and the platform fee does not apply. Adding credits is the simpler route to the rest of the catalog, but bringing your own keys stays available.

Is there a subscription or a minimum?

No. There are no plans at all — no subscription, no monthly fee, no minimum spend, and nothing to cancel. You pay only for what you run beyond the free allowance, from credits you chose to add.

What is MindsHub Air?

MindsHub Air is an alias, not a fixed model name. We keep it pointed at the model that strikes the best balance of cost, speed and intelligence, and which model that is can change without notice as better options ship — so you stay on that pick without changing anything. It is the model the 5M free monthly tokens apply to; past the allowance you keep using it at its own rates on this page, listed like every other model. Two limits are worth knowing before you lean on it: Air carries no web search and no reasoning-effort control, and it is not a stand-in for a frontier model on the hardest work — the rankings at mindshub.ai/models are where you pick one of those, and Air has its own row there.

What happens when the free allowance runs out?

It refreshes at the start of each month. Until it does, you have two ways to keep working: add credits and carry on at the published rates plus the 5% platform fee, or bring your own provider API key and pay that provider directly. Every token type counts toward the 5M — input, output, cached reads and cache writes.

Is there a platform fee?

Yes. The published rates mirror each provider's own prices, and a 5% platform fee is added to your bill on top. The fee is the same whether usage comes from the MindsHub Cowork workspace or from the Unified Inference API, and it never applies to the free monthly allowance or to bring-your-own-key usage. Volume customers can arrange a custom platform fee and volume discounts — reach the team through the contact form at mindshub.ai/contact.

Do Cowork and Unified Inference share this pricing?

Yes — one rate card and one balance. Whether a request comes from the MindsHub Cowork workspace or from your own code through Unified Inference (api.mindshub.ai), it bills at the same per-model rates plus the 5% platform fee, against the same prepaid balance. Nothing on this page is specific to one product.

Which models are available?

MindsHub Air is the model the free monthly allowance applies to. Adding credits opens the rest: the Claude family from Anthropic, the GPT family from OpenAI, Gemini from Google, Grok from SpaceXAI, Muse Spark from Meta, and open models including Kimi, DeepSeek, Qwen, and GLM. Exact versions move as providers ship releases, so the table on this page is authoritative: it lists every model with its provider, its input and output price per 1M tokens, and its failover order.

Can I see what a task cost?

Yes. Usage is visible in your account, and the rate card on this page shows exactly what each model charges, so you can work out the cost of a task before you run another like it.

How do I get support?

While you are on the free allowance, support is the MindsHub Discord community at mindshub.ai/discord — the team answers there alongside other users. Once you have added credits you can also open a support ticket through the form at mindshub.ai/support and get a tracked reply from the team. Discord stays open either way.

Can I pay with cryptocurrency?

Yes. Standard prepaid top-ups are paid by card. Volume customers may also settle their bill in cryptocurrency, arranged with our team alongside a custom platform fee and volume discounts. To set this up, contact us through the form at mindshub.ai/contact and mention cryptocurrency payments.

How do models stay current?

Models are referenced by family rather than pinned to versions. When a provider ships a new version, the Model Router resolves to it automatically. Most models also have a designated fallback the router switches to if a provider has an outage — this page shows the failover order per model.

What is MindsHub?

MindsHub is two products on one catalog and one balance. Unified Inference is the API: every major model behind one endpoint, in the OpenAI and Anthropic request formats. Cowork is the agent workspace on top of it — delegate entire projects and collect finished, shareable results.

Can I run MindsHub myself?

Yes. MindsHub is open source and runs anywhere — your own computer or your VPC — with your own model endpoints and API keys. The rates on this page cover the hosted MindsHub service at mindshub.ai, where the Model Router manages models for you.

Which open-source agents power MindsHub?

MindsHub Cowork runs on two interchangeable open-source agent harnesses — Anton, the default, and Hermes. Switch between them in a click and keep your workspace, data, and artifacts.